Senior Staff Security Engineer, AI Safeguards - Moveworks
Remote • Mountain View, CALIFORNIA, United States • Full-time
Posted 15h ago
Job Location
Mountain View, CALIFORNIA, United States
Tech Stack
Remote Work Policy
Fully remote
Employment Type
Full-time
Categories
AI Engineer
About the job
Moveworks is seeking a Senior Staff Security Engineer to focus on AI Safeguards. In this role, you will be responsible for maintaining an up-to-date map of ML usage and safeguards across our products, and leading the security design for new ML features. You will actively test the boundaries of our safeguards against various attacks like jailbreaks, prompt injection, and toxicity, translating these findings into evaluation cases that are applied to every model, prompt, and safeguard change. Additionally, you will collaborate with ML engineers to design and build robust mitigations that remain effective even when detection fails, and define and report on key security metrics to demonstrate the efficacy of our safeguards. You will also stay abreast of the latest AI attack research, incorporating new techniques into our testing and mitigation strategies.
Responsibilities
- Maintain ML usage and safeguards map across products
- Lead security design for new ML features
- Test LLM safeguards against attacks (jailbreaks, prompt injection, toxicity)
- Develop evaluation cases for model, prompt, and safeguard changes
- Design and build mitigations with ML engineers
- Define and report on security metrics for safeguards
- Review changes that could weaken safeguards
- Track AI attack research and develop tests and mitigations
Requirements
- 8+ years in security engineering
- 2+ years focused on AI or LLM security
- Hands-on adversarial experience (red teaming, penetration testing, vulnerability research) applied to LLM systems
- Deep knowledge of LLM attacks and defenses (e.g., OWASP Top 10 for LLM Applications, MITRE ATLAS)
- Ability to identify risks, develop mitigation plans, and implement fixes
- Strong coding skills in Python or Go
- Record of shipping production code
- Sufficient ML fundamentals to build evals and judge classifier efficacy